# A Real-Time Trajectory Optimization Method for Hypersonic Vehicles Based on a Deep Neural Network

^{1}

^{2}

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Materials and Methods

#### 2.1. Three-DOF Dynamic Model Development

#### 2.2. Problem Statement

#### 2.2.1. Dynamic Pressure Constraint

#### 2.2.2. Heat Flow Constraint

#### 2.2.3. Overload Constraint

#### 2.3. Research Ideas

## 3. Sample Data Generation Method Based on Chebyshev Pseudo-Spectral Method

#### 3.1. Chebyshev Pseudo-Spectral Method

#### 3.2. Training Data Generation

## 4. Neural Network Design and Training

Algorithm 1 Imitation learning |

1: Initialize network weighting values $\omega $ and $\alpha $ 2: Set $lr=0.0001,n\_epochs=30,batch\_size=256$ 3: for $epoch=1,n\_epochs$ do 4: for $batch\_index=1,n\_batches$ do 5: obtain the optimal sequence of pseudo-spectral method ballistic $\left[s,a\right]$ 6: $net\_in=\left[{s}_{0},{s}_{f},{s}_{current}\left],net\_out=\right[\alpha ,\sigma \right]$ data feature extraction and normalization 7: update network parameters using Adam algorithm: $loss=\frac{1}{n}{\displaystyle \sum _{i=1}^{n}}{[f\left({x}_{i}\right)-{y}_{i}]}^{2}$ 8: end for 9: Randomly generate a ballistic path by pseudo-spectral method $\left[{s}_{1},{a}_{1}\right]$ set up data buffering ℜ 10: $if$$dx\_angle{0.1}^{\xb0}$ do 11: $\mathrm{use}\mathrm{neural}\mathrm{network},\mathrm{input}\left[{s}_{0},{s}_{f},{s}_{current}\right]$$,\mathrm{output}\left[\alpha ,\sigma \right]$ 12: $\mathrm{put}\left[\alpha ,\sigma \right]\mathrm{into}\mathrm{environment()},\mathrm{obtain}{s}_{current+1}$ 13: $\mathrm{store}\mathrm{samples}\left[{s}_{0},{s}_{f},{s}_{current}\right]$, $\left[\alpha ,\sigma \right]$ to $\Re $, update ${s}_{current}$ 14: end |

## 5. Simulations and Result Analysis

^{2}. The CAV-H had a high maximum lift-to-drag ratio of E* = 3.24, and the corresponding lift coefficient ${C}_{L}^{*}$ was 0.45. The pneumatic reference area was ${\mathrm{s}}_{\mathrm{ref}}$= 0.8. The gravitational acceleration was${g}_{0}$= 9.8 m/s

^{2}, and the Earth radius was considered to be ${R}_{0}$= 6378 km.

#### 5.1. Generation of the Training Data

#### 5.2. Training Process of the DNN

#### 5.3. Random Single Trajectory Error Analysis

#### 5.4. Validation with Vehicle Dynamics Model

#### 5.5. Monte Carlo Simulation Verification

## 6. Conclusions

## Author Contributions

## Funding

## Institutional Review Board Statement

## Informed Consent Statement

## Data Availability Statement

## Acknowledgments

## Conflicts of Interest

## References

- Stryk, O.V.; Bulirsch, R. Direct and indirect methods for trajectory optimization. Ann. Oper. Res.
**1992**, 37, 357–373. [Google Scholar] [CrossRef] - Mansell, J.R.; Grant, M.J. Adaptive Continuation Strategy for Indirect Hypersonic Trajectory Optimization. J. Spacecr. Rocket.
**2018**, 55, 818–828. [Google Scholar] [CrossRef] - Grant, M.J.; Braun, R.D. Rapid Indirect Trajectory Optimization for Conceptual Design of Hypersonic Missions. J. Spacecr. Rocket.
**2015**, 52, 177–182. [Google Scholar] [CrossRef] - Tang, G.; Jiang, F.; Li, J. Fuel-Optimal Low-Thrust Trajectory Optimization Using Indirect Method and Successive Convex Programming. IEEE Trans. Aerosp. Electron. Syst.
**2018**, 54, 2053–2066. [Google Scholar] [CrossRef] - Taheri, E.; Kolmanovsky, I.; Atkins, E. Enhanced Smoothing Technique for Indirect Optimization of Minimum-Fuel Low-Thrust Trajectories. J. Guid. Control. Dyn.
**2016**, 39, 2500–2511. [Google Scholar] [CrossRef] [Green Version] - Wall, B.J.; Conway, B.A. Shape-Based Approach to Low-Thrust Rendezvous Trajectory Design. J. Guid. Control. Dyn.
**2009**, 32, 95. [Google Scholar] [CrossRef] - Subbarao, K.; Shippey, B.M. Hybrid Genetic Algorithm Collocation Method for Trajectory Optimization. J. Guid. Control. Dyn.
**2009**, 32, 1396–1403. [Google Scholar] [CrossRef] - Yang, S.; Cui, T.; Hao, X.; Yu, D. Trajectory optimization for a ramjet-powered vehicle in ascent phase via the Gauss pseudospectral method. Aerosp. Sci. Technol.
**2017**, 67, 88–95. [Google Scholar] [CrossRef] - Patterson, M.A.; Rao, A.V. GPOPS-II: A MATLAB Software for Solving Multiple-Phase Optimal Control Problems Using hp-Adaptive Gaussian Quadrature Collocation Methods and Sparse Nonlinear Programming. ACM Trans. Math. Softw.
**2010**, 41, 1–37. [Google Scholar] [CrossRef] [Green Version] - Lekkas, A.M.; Roald, A.L.; Breivik, M. Online Path Planning for Surface Vehicles Exposed to Unknown Ocean Currents Using Pseudospectral Optimal Control. IFAC-Pap. OnLine
**2016**, 49, 1–7. [Google Scholar] [CrossRef] - Bittner, M.; Fisch, F.; Holzapfel, F. A Multi-Model Gauss Pseudospectral Optimization Method for Aircraft Trajectories. In Proceedings of the AIAA Atmospheric Flight Mechanics Conference, Minneapolis, MN, USA, 13–16 August 2012. [Google Scholar]
- Antony, T.; Grant, M.J. Rapid indirect trajectory optimization on highly parallel computing architectures. J. Spacecr. Rocket.
**2017**, 54, 1081–1091. [Google Scholar] [CrossRef] [Green Version] - Wang, J.; Liang, H.; Qi, Z.; Ye, D. Mapped Chebyshev pseudospectral methods for optimal trajectory planning of differentially flat hypersonic vehicle systems. Aerosp. Sci. Technol.
**2019**, 89, 420–430. [Google Scholar] [CrossRef] - Wang, Z.; Mcdonald, S.T. Convex relaxation for optimal rendezvous of unmanned aerial and ground vehicles. Aerosp. Sci. Technol.
**2020**, 99, 105756.1–105756.19. [Google Scholar] [CrossRef] - Li, Y.; Pang, B.; Wei, C.; Cui, N.; Liu, Y. Online trajectory optimization for power system fault of launch vehicles via convex programming. Aerosp. Sci. Technol.
**2020**, 98, 105682. [Google Scholar] [CrossRef] - Zhang, Z.; Li, J.; Wang, J. Sequential convex programming for nonlinear optimal control problems in UAV path planning. Aerosp. Sci. Technol.
**2018**, 76, 280–290. [Google Scholar] [CrossRef] - Wang, Z. Optimal trajectories and normal load analysis of hypersonic glide vehicles via convex optimization. Aerosp. Sci. Technol.
**2019**, 87, 357–368. [Google Scholar] [CrossRef] - Wang, F.; Yang, S.; Xiong, F.F.; Lin, Q.; Song, J. Robust trajectory optimization using polynomial chaos and convex optimization. Aerosp. Sci. Technol.
**2019**, 92, 314–325. [Google Scholar] [CrossRef] - Han, H.; Qiao, D.; Chen, H.; Li, X. Rapid planning for aerocapture trajectory via convex optimization. Aerosp. Sci. Technol.
**2019**, 84, 763–775. [Google Scholar] [CrossRef] - Blackmore, L.; Scharf, D.P. Minimum-Landing-Error Powered-Descent Guidance for Mars Landing Using Convex Optimization. J. Guid. Control. Dyn.
**2010**, 33, 1161–1171. [Google Scholar] [CrossRef] [Green Version] - Wang, Z.; Lu, Y. Improved sequential convex programming algorithms for entry trajectory optimization. J. Spacecr. Rocket.
**2020**, 57, 1373–1386. [Google Scholar] [CrossRef] - De Bruijn, F.J.; Theil, S.; Choukroun, D.; Gill, E.K.A. Geostationary Satellite Station-Keeping Using Convex Optimization. J. Guid. Control. Dyn.
**2015**, 39, 605–616. [Google Scholar] [CrossRef] [Green Version] - Lin, B.; Carpenter, M.; Weck, O.D. Simultaneous Vehicle and Trajectory Design using Convex Optimization. In Proceedings of the AIAA Scitech 2020 Forum, Orlando, FL, USA, 6–10 January 2020. [Google Scholar]
- Thuruthel, T.G.; Shih, B.; Laschi, C.; Tolley, M.T. Soft robot perception using embedded soft sensors and recurrent neural networks. Sci. Robot.
**2019**, 4, eaav1488. [Google Scholar] [CrossRef] [PubMed] - Messaoud, K.; Yahiaoui, I.; Verroust-Blondet, A.; Nashashibi, F. Attention Based Vehicle Trajectory Prediction. IEEE Trans. Intell. Veh.
**2021**, 6, 175–185. [Google Scholar] [CrossRef] - Messaoud, K.; Yahiaoui, I.; Verroust-Blondet, A.; Nashashibi, F. Non-local Social Pooling for Vehicle Trajectory Prediction. In Proceedings of the 2019 IEEE Intelligent Vehicles Symposium (IV), Paris, France, 9–12 June 2019; pp. 975–980. [Google Scholar] [CrossRef] [Green Version]
- Sy, A.; Jian, L.A.; Lin, C.B. Low-thrust spacecraft trajectory optimization via a DNN-based method. Adv. Space Res.
**2020**, 66, 1635–1646. [Google Scholar] - Furfaro, R.; Linares, R. Deep Learning for Autonomous Lunar Landing. In Proceedings of the AAS/AIAA Astrodynamics Specialist Conference, Snowbird, UT, USA, 19–23 August 2018. [Google Scholar]
- Shi, Y.; Wang, Z. A Deep Learning-Based Approach to Real-Time Trajectory Optimization for Hypersonic Vehicles. In Proceedings of the AIAA Scitech 2020 Forum, Orlando, FL, USA, 6–10 January 2020. [Google Scholar]

**Figure 3.**State of training data. (

**a**) Height–time curve; (

**b**) Longitude–time curve; (

**c**) Latitude–time curve; (

**d**) velocity–time curve.

**Figure 4.**Control of training data. (

**a**) Generalized lift coefficient–time curve; (

**b**) Heeling Angle–time curve.

**Figure 5.**Training results of deep neural network. (

**a**) The training loss for sigmoid activation function epochs; (

**b**) the training loss for the ReLU activation function; (

**c**) the test loss for the sigmoid activation function; (

**d**) the test loss for the ReLU activation function.

**Figure 6.**Comparison of the predicted and expected values of the generalized coefficient of the lift.

**Figure 8.**Comparison of the predicted and expected values of the generalized coefficient of the lift.

Parameter | Value Range |
---|---|

Initial height h_{0} | 41 km~46 km |

Initial longitude θ_{0} | −2°~2° |

Initial latitude φ_{0} | −2°~2° |

Initial velocity V_{0} | 5300 m/s |

Initial track angle γ_{0} | 0° |

Initial course angle ψ_{0} | 90° |

Final longitude θ_{f} | 38°~42° |

Final latitude φ_{f} | 18°~22° |

Parameter | $\dot{\mathit{Q}}{\left(\frac{\mathbf{kW}}{{\mathit{m}}^{2}}\right)}_{\mathit{m}\mathit{a}\mathit{x}}$ | $\overline{\mathit{q}}{\left(\mathbf{kPa}\right)}_{\mathit{m}\mathit{a}\mathit{x}}$ | $\mathit{n}{\left({\mathit{g}}_{0}\right)}_{\mathit{m}\mathit{a}\mathit{x}}$ | Generalized Lift Coefficient | Heeling Angle (°) |
---|---|---|---|---|---|

Value | 2000 | 500 | 3 | $0\le \lambda \le 2$ | $-80\le \sigma \le 80$ |

Actual Vehicle Position | Predicted Vehicle Position | Position Error | |
---|---|---|---|

Altitude (m) | 30,151 | 30,940 | 789 |

Longitude (°) | 34.84 | 34.74 | 0.10 |

Latitude (°) | 18.16 | 18.19 | 0.03 |

Velocity (m/s) | 2267 | 2271 | 4 |

The Absolute Terminal Longitude | The Absolute Terminal Latitude | The Absolute Terminal Range Angle | |
---|---|---|---|

error (°) | 0.042 | 0.125 | 0.126 |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Wang, J.; Wu, Y.; Liu, M.; Yang, M.; Liang, H.
A Real-Time Trajectory Optimization Method for Hypersonic Vehicles Based on a Deep Neural Network. *Aerospace* **2022**, *9*, 188.
https://doi.org/10.3390/aerospace9040188

**AMA Style**

Wang J, Wu Y, Liu M, Yang M, Liang H.
A Real-Time Trajectory Optimization Method for Hypersonic Vehicles Based on a Deep Neural Network. *Aerospace*. 2022; 9(4):188.
https://doi.org/10.3390/aerospace9040188

**Chicago/Turabian Style**

Wang, Jianying, Yuanpei Wu, Ming Liu, Ming Yang, and Haizhao Liang.
2022. "A Real-Time Trajectory Optimization Method for Hypersonic Vehicles Based on a Deep Neural Network" *Aerospace* 9, no. 4: 188.
https://doi.org/10.3390/aerospace9040188